A Strategy Selection Framework for Adaptive Prefetching in Data Visualization

Punit R. Doshi, Geraldine E. Rosario, Elke A. Rundensteiner, Matthew O. Ward

Abstract: Accessing data stored in persistent memory represents a bottleneck for current visual exploration applications. Semantic caching of frequent queries at the client-side along with prefetching can improve performance of such systems. However, a prefetching setup that only uses one prefetching strategy may be insufficient because (1) different users have different exploration patterns, and (2) a user's pattern may be changing within the same session. To solve this, existing research focuses on refining a single prefetching strategy. We, on the other hand, now propose a framework wherein prefetching strategies are adaptively selected over time across and within one user session. This work is the first to study adaptive prefetching in the context of visual data exploration. Specifically, we have implemented our proposed approach within XmdvTool, a freeware visualization system for multivariate data, and evaluated it using real user traces. Our results confirm that our approach improves system performance by dynamically selecting the most appropriate combination of prefetching strategies that adapts to the user's changing patterns.